Merge pull request 'fix: [bug] IR 覆盖率回归：Layer B 从 92.6% 降至 63% + Layer A 新 schema 错误 - 来自 #18 - Closes #57' (#59) from dev/issue-57-fix-coverage-regression into main

skills/ir_generation_skill/step1_semantic_index.py

@@ -880,7 +880,7 @@ def run_ensemble_semantic_index(doc: dict) -> dict:
             if v:
                 print(f"    {k}: {len(v)} 个问题")
 
-        # Feedback retry: re-run with coverage feedback (up to 2 retries)
+        # Feedback retry: re-run with coverage feedback (up to 2 retries, quality-gated)
         retry_count = 0
         while retry_count < 2:
             feedback = _build_coverage_feedback(gaps)
@@ -889,6 +889,10 @@ def run_ensemble_semantic_index(doc: dict) -> dict:
             retry_count += 1
             print(f"\n  覆盖反馈重试 #{retry_count} (feedback长度={len(feedback)}字符)...", flush=True)
             try:
+                # record pre-retry coverage to gate quality
+                pre_warnings = len(gaps.get("coverage_warnings", []))
+                pre_missing_rows = len(gaps.get("missing_table_rows", []))
+
                 retry_prompt = build_prompt(doc, feedback, all_paths)
                 print(f"  重试 prompt 长度: {len(retry_prompt)} 字符", flush=True)
                 retry_result = call_llm(retry_prompt, max_retries=1, temperature=0.3)
@@ -902,15 +906,28 @@ def run_ensemble_semantic_index(doc: dict) -> dict:
                             if src.get("section"):
                                 retry_sections.add(src["section"])
                     print(f"  重试新增 sections: {sorted(retry_sections)}", flush=True)
-                    semantic_indices.append(retry_result)
+                    # Quality gate: only include retry if it improves coverage
-                    merged = ensemble_merge(semantic_indices)
+                    trial_indices = semantic_indices + [retry_result]
-                    merged["ensemble_temperatures"] = list(temperatures) + [f"feedback_retry_{retry_count}"]
+                    trial_merged = ensemble_merge(trial_indices)
-                    passed, gaps = _quick_validate(merged, doc, all_paths)
+                    trial_passed, trial_gaps = _quick_validate(trial_merged, doc, all_paths)
-                    merged["validation_passed"] = passed
+                    trial_warnings = len(trial_gaps.get("coverage_warnings", []))
-                    merged["validation_gaps"] = {
+                    trial_missing = len(trial_gaps.get("missing_table_rows", []))
-                        k: v for k, v in gaps.items() if v
+                    if trial_warnings < pre_warnings or trial_missing < pre_missing_rows:
-                    }
+                        semantic_indices.append(retry_result)
-                    print(f"  重试后验证: {'PASS' if passed else 'GAPS FOUND'}", flush=True)
+                        merged = trial_merged
+                        passed, gaps = trial_passed, trial_gaps
+                        merged["ensemble_temperatures"] = list(temperatures) + [f"feedback_retry_{retry_count}"]
+                        merged["validation_passed"] = passed
+                        merged["validation_gaps"] = {
+                            k: v for k, v in gaps.items() if v
+                        }
+                        print(f"  重试后验证 (已采纳): {'PASS' if passed else 'GAPS FOUND'} "
+                              f"(warnings {pre_warnings}→{trial_warnings}, "
+                              f"missing_rows {pre_missing_rows}→{trial_missing})", flush=True)
+                    else:
+                        print(f"  重试结果未提升覆盖率，丢弃 "
+                              f"(warnings {pre_warnings}→{trial_warnings}, "
+                              f"missing_rows {pre_missing_rows}→{trial_missing})", flush=True)
             except Exception as e:
                 print(f"  覆盖反馈重试失败: {e}", flush=True)
                 import traceback

skills/ir_generation_skill/step3_merge_and_audit.py

@@ -170,8 +170,11 @@ def _normalize_rule(rule: dict) -> dict:
         }]
 
     # Ensure table/text sources have a section field (defensive against LLM omission)
+    # Also normalize invalid source types (LLM hallucinations like function_unit_description)
     sources = rule.get("sources", [])
     if sources:
+        valid_types = {"table", "text", "logic_tree"}
+
         # try to infer a default section from sibling sources or the rule path
         default_section = ""
         for s in sources:
@@ -186,6 +189,10 @@ def _normalize_rule(rule: dict) -> dict:
 
         for src in sources:
             stype = src.get("type", "")
+            # Normalize invalid source types to "text"
+            if stype and stype not in valid_types:
+                src["type"] = "text"
+                stype = "text"
             if stype in ("table", "text"):
                 if not src.get("section"):
                     src["section"] = default_section

skills/ir_generation_skill/tests/test_step3.py

@@ -511,3 +511,18 @@ class TestNormalizeRule:
         }
         normalized = _normalize_rule(rule)
         assert "section" not in normalized["sources"][0]
+
+    def test_normalize_source_invalid_type(self):
+        """Invalid source types (LLM hallucinations) are normalized to text."""
+        rule = {
+            "trigger": {"conditions": [{"signal": "x", "operator": "==", "value": "1"}]},
+            "sources": [
+                {"type": "function_unit_description", "text_snippet": "desc",
+                 "section": "3.1 功能"},
+                {"type": "unknown_type", "text_snippet": "also invalid"},
+            ],
+        }
+        normalized = _normalize_rule(rule)
+        assert normalized["sources"][0]["type"] == "text"
+        assert normalized["sources"][1]["type"] == "text"
+        assert normalized["sources"][0]["section"] == "3.1 功能"